rsnt_translations
56,430
edits
Multi-Instance GPU: Difference between revisions
Jump to navigation
Jump to search
no edit summary
No edit summary |
No edit summary |
||
| Line 16: | Line 16: | ||
=Limitations = <!--T:4--> | =Limitations = <!--T:4--> | ||
[https://docs.nvidia.com/datacenter/tesla/mig-user-guide/index.html#app-considerations | [https://docs.nvidia.com/datacenter/tesla/mig-user-guide/index.html#app-considerations The MIG technology does not support] [https://developer.nvidia.com/docs/drive/drive-os/6.0.8.1/public/drive-os-linux-sdk/common/topics/nvsci_nvsciipc/Inter-ProcessCommunication1.html CUDA Inter-Process Communication (IPC)], which optimizes data transfers between GPUs over NVLink and NVSwitch. | ||
This limitation also | This limitation also prevents efficient communication between instances of a full GPU. | ||
Consequently, launching an executable on more than one | Consequently, launching an executable on more than one instance at a time <b>does not</b> improve performance and should be avoided. | ||
<!--T:5--> | <!--T:5--> | ||
GPU jobs requiring many CPU cores may also require a full GPU instead of | GPU jobs requiring many CPU cores may also require a full GPU instead of an instance. The maximum number of CPU cores per instance depends on [[Allocations_and_compute_scheduling#Ratios_in_bundles|the number of cores per full GPU]] and on the configured [https://docs.nvidia.com/datacenter/tesla/mig-user-guide/index.html#a100-profiles MIG profiles]. Both factors may vary between clusters and also between GPU nodes in a cluster. | ||
= Available configurations = <!--T:6--> | = Available configurations = <!--T:6--> | ||